[Sched] Skip MemOp with unknown size when clustering

[wangpc-pp]

In #83875, we changed the type of Width to LocationSize. To get
the clsuter bytes, we use LocationSize::getValue() to calculate
the value.

But when Width is an unknown size LocationSize, an assertion
"Getting value from an unknown LocationSize!" will be triggered.

This patch simply skips MemOp with unknown size to fix this issue
and keep the logic the same as before.

This issue was found when implementing software pipeliner for
RISC-V in #117546. The pipeliner may clone some memory operations
with BeforeOrAfterPointer size.

[nikic]

LGTM. I assume that this is going to be tested by the following RISCV PR.

[wangpc-pp]

Code that will generate unknown size LocationSize:

llvm-project/llvm/lib/CodeGen/ModuloSchedule.cpp

Lines 962 to 994 in c1ad064

	/// Update the memory operand with a new offset when the pipeliner
	/// generates a new copy of the instruction that refers to a
	/// different memory location.
	void ModuloScheduleExpander::updateMemOperands(MachineInstr &NewMI,
	MachineInstr &OldMI,
	unsigned Num) {
	if (Num == 0)
	return;
	// If the instruction has memory operands, then adjust the offset
	// when the instruction appears in different stages.
	if (NewMI.memoperands_empty())
	return;
	SmallVector<MachineMemOperand *, 2> NewMMOs;
	for (MachineMemOperand *MMO : NewMI.memoperands()) {
	// TODO: Figure out whether isAtomic is really necessary (see D57601).
	if (MMO->isVolatile() || MMO->isAtomic() ||
	(MMO->isInvariant() && MMO->isDereferenceable()) ||
	(!MMO->getValue())) {
	NewMMOs.push_back(MMO);
	continue;
	}
	unsigned Delta;
	if (Num != UINT_MAX && computeDelta(OldMI, Delta)) {
	int64_t AdjOffset = Delta * Num;
	NewMMOs.push_back(
	MF.getMachineMemOperand(MMO, AdjOffset, MMO->getSize()));
	} else {
	NewMMOs.push_back(MF.getMachineMemOperand(
	MMO, 0, LocationSize::beforeOrAfterPointer()));
	}
	}
	NewMI.setMemRefs(MF, NewMMOs);
	}

[arsenm]

I'd somewhat expect this to just collect the unknown width and let the users figure out what to do with it? Would that require widespread changes?

[wangpc-pp]

Yeah, I think we need to modify clusterNeighboringMemOps and other places as well. That can be a future work if you agree.